1
00:00:06,740 --> 00:00:11,420
Hello. My name is Yondon. Thanks everyone for coming tonight. I'm

2
00:00:11,420 --> 00:00:16,540
going to go through this really fast. My main intent here is

3
00:00:16,540 --> 00:00:20,420
less so to give a long, lengthy talk and more so to kind of seed

4
00:00:20,420 --> 00:00:23,000
a couple topics that I think are interesting for conversation later

5
00:00:23,000 --> 00:00:25,700
in the night. So let's get going.

6
00:00:26,640 --> 00:00:29,260
So I'm going to focus on open source software

7
00:00:29,260 --> 00:00:31,080
and what's happening in open source software today.

8
00:00:31,800 --> 00:00:33,240
Hopefully it becomes a little bit more clear

9
00:00:33,240 --> 00:00:35,940
why I've chosen this topic, and happy to talk more

10
00:00:35,940 --> 00:00:36,860
about that at the end of the talk.

11
00:00:37,340 --> 00:00:39,980
But I think there's a very clear dilemma in the open source

12
00:00:39,980 --> 00:00:40,980
software communities today.

13
00:00:41,180 --> 00:00:43,180
And it looks something like this, which is,

14
00:00:43,300 --> 00:00:45,560
you have a human sending a fleet of agents

15
00:00:45,560 --> 00:00:48,000
that are opening gigantic PRs left and right

16
00:00:48,000 --> 00:00:49,380
on popular software projects.

17
00:00:49,380 --> 00:00:52,020
And they ostensibly look good.

18
00:00:52,300 --> 00:00:53,240
The code all checks out.

19
00:00:53,440 --> 00:00:55,120
But a human maintainer on the other end

20
00:00:55,120 --> 00:00:58,040
is super stressed because there are now tons of demands

21
00:00:58,040 --> 00:00:58,780
for their attention.

22
00:00:59,180 --> 00:01:01,180
And even though the PRs look pretty good at face value,

23
00:01:01,580 --> 00:01:03,260
oftentimes they are very narrow.

24
00:01:03,420 --> 00:01:04,220
They lack context.

25
00:01:04,440 --> 00:01:06,600
They miss requirements that are not stated explicitly.

26
00:01:07,100 --> 00:01:08,720
And ultimately, nothing can really be merged.

27
00:01:10,340 --> 00:01:14,120
So in response to this, what we've seen in some areas

28
00:01:14,120 --> 00:01:16,620
is projects just saying, we don't need your contributions

29
00:01:16,620 --> 00:01:16,980
anymore.

30
00:01:16,980 --> 00:01:18,680
We don't want your contributions anymore.

31
00:01:19,380 --> 00:01:21,540
And it's hard to fault them for this policy.

32
00:01:21,860 --> 00:01:25,340
For example, tldraw, a prominent whiteboarding software

33
00:01:25,340 --> 00:01:28,120
earlier this year, said that they're automatically closing PRs going

34
00:01:28,120 --> 00:01:28,360
forward.

35
00:01:28,520 --> 00:01:30,160
They don't want external contributors anymore.

36
00:01:30,200 --> 00:01:33,440
And it doesn't have to do with the external contributors being malicious.

37
00:01:33,620 --> 00:01:36,340
It's just the state of the ecosystem is untenable for these projects

38
00:01:36,340 --> 00:01:36,640
anymore.

39
00:01:36,640 --> 00:01:40,740
Because why would you go through that entire demand on your attention

40
00:01:40,740 --> 00:01:41,800
when you can just

41
00:01:41,800 --> 00:01:43,120
have your own fleet of agents?

42
00:01:43,240 --> 00:01:46,100
You tell them the issues and the fleet of agents does the work for

43
00:01:46,100 --> 00:01:46,320
you.

44
00:01:46,600 --> 00:01:47,360
Isn't that so much better?

45
00:01:47,420 --> 00:01:48,420
You don't have to deal with contributors.

46
00:01:49,560 --> 00:01:51,580
And it's unclear where this all goes.

47
00:01:51,900 --> 00:01:55,900
But one possible state of the world is that we just stopped seeing

48
00:01:55,900 --> 00:01:56,900
open contribution at

49
00:01:56,900 --> 00:01:57,140
scale.

50
00:01:57,660 --> 00:02:01,980
This was maybe going to be looked back on in history as a blip where

51
00:02:01,980 --> 00:02:03,080
for maybe like a couple

52
00:02:03,080 --> 00:02:06,180
decade period, had open contribution at scale on the internet.

53
00:02:06,180 --> 00:02:08,080
And it's not really going to be a thing anymore.

54
00:02:08,300 --> 00:02:12,220
this is a post sharing the sentiment from Mitchell Hashimoto, previous

55
00:02:12,220 --> 00:02:13,220
co-founder of HashiCorp

56
00:02:13,220 --> 00:02:16,680
and then also maintainer of a really prominent open source project,

57
00:02:16,740 --> 00:02:17,020
Ghostty.

58
00:02:18,760 --> 00:02:20,780
So the question is, is this fine?

59
00:02:21,820 --> 00:02:22,700
So maybe.

60
00:02:23,140 --> 00:02:26,380
I mean, a centralized software factory with agents can still produce

61
00:02:26,380 --> 00:02:27,580
good, useful software.

62
00:02:27,980 --> 00:02:31,220
But the question is, is the software ultimately the only thing that

63
00:02:31,220 --> 00:02:31,880
we cared about in the first

64
00:02:31,880 --> 00:02:32,100
place?

65
00:02:32,100 --> 00:02:36,260
And I think there's an argument that for anyone that has participated

66
00:02:36,260 --> 00:02:37,380
in open source, open source

67
00:02:37,380 --> 00:02:38,640
kind of has two products.

68
00:02:38,900 --> 00:02:42,100
It has the software that's produced, but it also has this maybe

69
00:02:42,100 --> 00:02:43,320
side effect, which is this

70
00:02:43,320 --> 00:02:45,960
community of shared trust and understanding, this sort of like knowledge

71
00:02:45,960 --> 00:02:47,040
commons that's

72
00:02:47,040 --> 00:02:48,940
built up from diverse participants.

73
00:02:49,400 --> 00:02:51,860
And that was like the second product of open source.

74
00:02:52,080 --> 00:02:54,140
And that's something that might go away.

75
00:02:54,280 --> 00:02:57,400
So is that something that's lost along the way in this process?

76
00:02:58,060 --> 00:03:00,680
So a natural question is, is there another way?

77
00:03:00,680 --> 00:03:04,420
Well, a natural response is, well, why don't we have agents do the

78
00:03:04,420 --> 00:03:05,140
reviewing so that you

79
00:03:05,140 --> 00:03:05,900
can keep the contributors?

80
00:03:06,200 --> 00:03:09,340
And then you just have agents on the other side doing the maintenance

81
00:03:09,340 --> 00:03:09,600
work.

82
00:03:10,760 --> 00:03:13,620
But I think naively done, this doesn't really completely solve the

83
00:03:13,620 --> 00:03:14,700
problem either, because

84
00:03:14,700 --> 00:03:16,660
you're just swapping one scarce resource for another.

85
00:03:17,160 --> 00:03:20,560
You're swapping scarce human attention for scarce AI context and

86
00:03:20,560 --> 00:03:20,820
tokens.

87
00:03:21,200 --> 00:03:23,100
You're not making the resource problem go away.

88
00:03:23,260 --> 00:03:24,780
You're just changing what resource is scarce.

89
00:03:24,780 --> 00:03:27,780
And ultimately, you can still end up with an overwhelming number

90
00:03:27,780 --> 00:03:29,040
of things that need to

91
00:03:29,040 --> 00:03:32,380
be done by the agents, even if they are doing it on behalf of the

92
00:03:32,380 --> 00:03:32,600
humans.

93
00:03:33,180 --> 00:03:37,540
So kind of my view on this right now is that if by default you're

94
00:03:37,540 --> 00:03:38,900
saying that you have unbounded

95
00:03:38,900 --> 00:03:43,020
public writes, meaning posts, code, whatever, that means unbounded

96
00:03:43,020 --> 00:03:44,500
by default context and

97
00:03:44,500 --> 00:03:45,840
inference costs for a maintainer.

98
00:03:45,840 --> 00:03:50,200
And each write in this workspace basically is a draw on shared attention,

99
00:03:50,500 --> 00:03:51,180
which is a shared

100
00:03:51,180 --> 00:03:54,660
pooled resource within a community, whether it be of a human or

101
00:03:54,660 --> 00:03:54,920
an AI.

102
00:03:55,820 --> 00:03:59,860
And some people, like the engineer-minded person can say that, well,

103
00:03:59,980 --> 00:04:00,720
we can make the maintainer

104
00:04:00,800 --> 00:04:04,180
It's more efficient and save on costs. But I don't think this is

105
00:04:04,180 --> 00:04:06,940
a complete solve either because it doesn't address how the rules

106
00:04:06,940 --> 00:04:09,120
of the system incentivize the contributors to behave in the first

107
00:04:09,120 --> 00:04:09,340
place.

108
00:04:09,600 --> 00:04:12,760
And that is, on one hand, maybe partially a distributed systems

109
00:04:12,760 --> 00:04:15,760
question, but also a mechanism design question.

110
00:04:15,940 --> 00:04:19,380
So it just requires a different flavor of thinking than just treating

111
00:04:19,380 --> 00:04:22,180
it as a pure engineering problem for the maintainers themselves.

112
00:04:22,400 --> 00:04:23,760
So ideally you have solutions for both.

113
00:04:25,080 --> 00:04:28,480
So a quick rundown. We've been experimenting with this a little

114
00:04:28,480 --> 00:04:29,520
bit, as Eric mentioned.

115
00:04:29,700 --> 00:04:33,960
And I think it's more fun to focus on the first early steps here.

116
00:04:34,140 --> 00:04:36,920
And it's more fun to focus on the failure modes first, because now

117
00:04:36,920 --> 00:04:40,100
you get a sense of what happens when you just do things naively.

118
00:04:40,480 --> 00:04:44,280
So in Commons today, you can form an agent organization with maintainers

119
00:04:44,280 --> 00:04:47,120
and contributors, but there aren't really sophisticated rules yet.

120
00:04:47,240 --> 00:04:50,620
So what happens when you just let them do the thing? And it turns

121
00:04:50,620 --> 00:04:53,240
out you get a lot of quirky behavior that you wouldn't expect from

122
00:04:53,240 --> 00:04:53,500
humans.

123
00:04:53,500 --> 00:04:56,880
So for example, if a human ran into a blocker for a task, they'd

124
00:04:56,880 --> 00:04:58,980
probably be like, I'm blocked. That's it.

125
00:04:59,620 --> 00:05:03,100
In this case, there was a task with literally impossible acceptance

126
00:05:03,100 --> 00:05:08,280
criteria, and the agent kind of dutifully obeying

127
00:05:08,280 --> 00:05:08,780
its instructions,

128
00:05:09,280 --> 00:05:11,920
which included give regular status updates, was like, I'm just going

129
00:05:11,920 --> 00:05:14,840
to keep giving you status updates on why this task is impossible.

130
00:05:15,080 --> 00:05:18,760
So 800 messages later, it's still saying that, oh, this is impossible.

131
00:05:18,960 --> 00:05:19,780
It cannot be completed.

132
00:05:21,480 --> 00:05:23,980
Or you just have, like, lots of redundancy that doesn't really make

133
00:05:23,980 --> 00:05:24,240
sense.

134
00:05:24,680 --> 00:05:26,880
You kind of deploy a fleet of agents. They're all looking at the

135
00:05:26,880 --> 00:05:28,620
same workspace, and they're like, we should all do this thing.

136
00:05:28,680 --> 00:05:30,320
It's a good thing. I'm going to create this task, and I'm going

137
00:05:30,320 --> 00:05:30,720
to do the thing.

138
00:05:31,140 --> 00:05:34,560
And then they all do the same exact thing, and then that wasn't

139
00:05:34,560 --> 00:05:35,120
really helpful, right?

140
00:05:35,120 --> 00:05:38,040
We only needed to do it once. And all of this happened in a very

141
00:05:38,040 --> 00:05:38,660
short time window.

142
00:05:39,220 --> 00:05:43,080
Or my favorite one is a status update about staying silent, where

143
00:05:43,080 --> 00:05:46,080
the maintainer, seeing that these status updates aren't really useful,

144
00:05:46,460 --> 00:05:49,180
tells the rest of the contributors, you should stop, no further

145
00:05:49,180 --> 00:05:50,020
acknowledgement posts.

146
00:05:50,220 --> 00:05:53,600
And the contributor, three minutes later, says, I'm complying with

147
00:05:53,600 --> 00:05:57,000
your hold, and then proceeds to write three paragraphs with its

148
00:05:57,000 --> 00:05:58,860
status report about why it's complying with the hold.

149
00:05:59,760 --> 00:06:02,160
So I mention all of these because they're kind of amusing, just

150
00:06:02,160 --> 00:06:04,640
like failure modes, when you just try this for the first time.

151
00:06:04,760 --> 00:06:07,880
And obviously, you want more sophisticated roles. But two observations

152
00:06:07,880 --> 00:06:08,340
I'd share.

153
00:06:08,680 --> 00:06:12,480
One, unlike with human attention, we can actually quantify the cost

154
00:06:12,480 --> 00:06:16,120
of the system in all of these cases, in the form of the maintainer's

155
00:06:16,120 --> 00:06:17,160
cost burden for inference.

156
00:06:17,380 --> 00:06:20,180
So every single time there's basically spam in one of these workspaces,

157
00:06:20,180 --> 00:06:22,980
you can see it numerically with how much it costs to run a maintainer.

158
00:06:22,980 --> 00:06:26,240
So that's interesting. And then the second observation is that there's

159
00:06:26,240 --> 00:06:28,540
just this interesting tension between what's the right behavior

160
00:06:28,540 --> 00:06:29,860
and what you should be optimizing for.

161
00:06:30,160 --> 00:06:32,840
You could argue that the agent is just following instructions. It

162
00:06:32,840 --> 00:06:33,980
was giving status updates.

163
00:06:34,140 --> 00:06:36,780
It just happens to be the case that the status updates is adding

164
00:06:36,780 --> 00:06:38,400
noise to everyone else in this workspace.

165
00:06:39,940 --> 00:06:42,460
So this has motivated a bunch of, I think, interesting areas of

166
00:06:42,460 --> 00:06:42,880
investigation.

167
00:06:43,260 --> 00:06:45,660
I'm not going to try to go through all of them right now. But if

168
00:06:45,660 --> 00:06:47,660
you want to talk about these, I think these are interesting lines

169
00:06:47,660 --> 00:06:48,140
of research.

170
00:06:48,340 --> 00:06:51,520
But it ranges from, okay, well, maybe there should be participatory

171
00:06:51,520 --> 00:06:52,660
budgets for these agents.

172
00:06:52,660 --> 00:06:55,360
Maybe they should have to intelligently budget what they choose

173
00:06:55,360 --> 00:06:58,100
to post about and what they choose to contribute to so that there

174
00:06:58,100 --> 00:06:59,240
is a form of scarcity for them.

175
00:07:00,400 --> 00:07:04,560
Or a web of trust-esque reputation systems that are tied with scoped,

176
00:07:04,740 --> 00:07:07,040
granular capabilities and permissions that scale with that trust.

177
00:07:07,420 --> 00:07:10,280
And then the last one that I'll mention, so agent identity is in

178
00:07:10,280 --> 00:07:10,520
there too.

179
00:07:10,700 --> 00:07:13,640
The last one I mentioned since it kind of alludes to just this broader

180
00:07:13,640 --> 00:07:15,840
conversation of like what is happening and how do we understand

181
00:07:15,840 --> 00:07:19,500
these things is evals for multi-principal agent works.

182
00:07:19,500 --> 00:07:23,140
I emphasize multi-principal because what we're assuming here is

183
00:07:23,140 --> 00:07:25,880
that everyone has diverse, different private preferences that may

184
00:07:25,880 --> 00:07:27,060
or may not align with one another.

185
00:07:27,280 --> 00:07:30,420
And then I would also emphasize that it's important for these evals

186
00:07:30,420 --> 00:07:32,060
to be reproducible and open.

187
00:07:32,260 --> 00:07:35,180
Because I think the, in my opinion, the biggest thing holding back

188
00:07:35,180 --> 00:07:38,060
discourse about these topics is that you don't really have this

189
00:07:38,060 --> 00:07:40,380
level of reproducibility and openness and transparency.

190
00:07:40,380 --> 00:07:43,660
So no one is quite talking about the same scenario because no one

191
00:07:43,660 --> 00:07:45,420
knows what they're talking about.

192
00:07:45,580 --> 00:07:48,460
You only are kind of referring to some report that someone gave

193
00:07:48,460 --> 00:07:50,480
you and you don't truly know what was happening.

194
00:07:50,700 --> 00:07:52,960
So ideally you would know the full prompt, you would know the full

195
00:07:52,960 --> 00:07:53,840
harness configuration.

196
00:07:54,560 --> 00:07:57,000
So I think these evals will be really interesting for these types

197
00:07:57,000 --> 00:07:57,920
of organizational structures.

198
00:07:58,720 --> 00:08:02,400
So I'll end where I chose to focus on open source software, but

199
00:08:02,400 --> 00:08:05,680
I think there are going to be a lot of similarities in terms of

200
00:08:05,680 --> 00:08:08,500
the ideas and problems that pertain to other fields as well when

201
00:08:08,500 --> 00:08:10,800
it comes to open communities and collective work.

202
00:08:10,980 --> 00:08:13,980
So I think I'll just end this question, end with this question,

203
00:08:14,120 --> 00:08:16,680
which is like how do we generally think about agent-native institutions

204
00:08:16,680 --> 00:08:18,640
for achieving results, which obviously we care about.

205
00:08:18,640 --> 00:08:21,660
But some of these other things, which is preserving or creating

206
00:08:21,660 --> 00:08:24,400
new processes for shared understanding, attention, and trust, something

207
00:08:24,400 --> 00:08:27,820
that is fundamental to open source, that can support collective

208
00:08:27,820 --> 00:08:29,000
work in open communities broadly.

209
00:08:29,460 --> 00:08:31,120
So I think I'll just end with that question.

210
00:08:31,540 --> 00:08:34,740
And yeah, I'll take some questions, but also happy to pick up the

211
00:08:34,740 --> 00:08:36,380
conversation separately as well.

212
00:08:36,900 --> 00:08:37,200
Thanks.

213
00:08:49,740 --> 00:08:50,540
Yeah.

214
00:08:52,720 --> 00:08:58,100
You mentioned about the problem of open

215
00:08:58,100 --> 00:09:03,140
source repositories dealing with way too

216
00:09:03,140 --> 00:09:08,260
many contributors, and maybe you can like scale agentic review,

217
00:09:08,440 --> 00:09:10,320
but then it still can be very costly.

218
00:09:11,720 --> 00:09:17,460
And you mentioned that maybe like a limiting number of tokens

219
00:09:17,460 --> 00:09:21,680
could be a solution, or how many contributions one agent could do

220
00:09:21,680 --> 00:09:22,420
could be a solution.

221
00:09:23,120 --> 00:09:27,220
Have you seen other like mitigation factors that have been useful

222
00:09:27,220 --> 00:09:28,600
for open source?

223
00:09:28,780 --> 00:09:31,500
Because there are open source projects that are not doing what TL

224
00:09:31,500 --> 00:09:33,500
draw did, but still accepting lots of contributions.

225
00:09:34,360 --> 00:09:36,440
Have you seen good examples that are inspiring?

226
00:09:36,440 --> 00:09:39,800
Yeah, I think one interesting example that's happening live right

227
00:09:39,800 --> 00:09:44,460
now is I share that post from Mitchell Hashimoto, but he has this

228
00:09:44,460 --> 00:09:47,520
project called Vouch, where they launched it earlier this year.

229
00:09:48,520 --> 00:09:51,620
And it's nothing too fancy, but it basically is built off of this

230
00:09:51,620 --> 00:09:54,720
notion of trust lists that can be maintained per repo, and then

231
00:09:54,720 --> 00:09:59,380
the ability to publicly basically vouch for or denounce GitHub identity.

232
00:09:59,640 --> 00:10:02,000
And it was controversial at the time, right, because people said

233
00:10:02,000 --> 00:10:04,540
that, oh, you can denounce someone, this is just going to be used

234
00:10:04,540 --> 00:10:05,040
for gatekeeping.

235
00:10:05,040 --> 00:10:08,080
But in practice, I think what it's been used for is to experiment

236
00:10:08,080 --> 00:10:12,860
with, okay, we, as the inner circle of maintainers, generally have

237
00:10:12,860 --> 00:10:15,900
a vibe of like who's been useful in contributing stuff.

238
00:10:16,000 --> 00:10:18,640
So why don't we just make that explicit? It's happening already

239
00:10:18,640 --> 00:10:19,300
implicitly.

240
00:10:19,340 --> 00:10:22,300
So it's kind of a lie to say that there isn't already this trust

241
00:10:22,300 --> 00:10:25,740
system. So why not just make it super explicit so everyone can transparently

242
00:10:25,740 --> 00:10:28,560
see who is being vouched for and who is being publicly denounced.

243
00:10:28,560 --> 00:10:31,560
And then I think the idea or the hope for that project is that people

244
00:10:31,560 --> 00:10:33,480
would be able to share trust lists with one another.

245
00:10:33,620 --> 00:10:37,740
So if I see that like some sus guy shows up and just gave me a thousand

246
00:10:37,740 --> 00:10:41,140
drive by PRs and then left, well, maybe someone else wants to know

247
00:10:41,140 --> 00:10:41,540
about that.

248
00:10:41,680 --> 00:10:45,120
So I think there's definitely room to kind of learn and maybe extend

249
00:10:45,120 --> 00:10:45,940
some of those primitives.

250
00:10:46,960 --> 00:10:51,100
So I don't think any of these, the areas of investigation should

251
00:10:51,100 --> 00:10:53,240
be treated as just like let's start from a blank slate.

252
00:10:53,400 --> 00:10:55,220
Because I think there's a lot of good work happening already.

253
00:10:55,440 --> 00:10:58,760
I think the main question is, you know, how does that get incorporated

254
00:10:58,760 --> 00:10:59,240
writ large?

255
00:10:59,460 --> 00:11:02,560
Is it a one size fits all solution or should we be thinking about

256
00:11:02,560 --> 00:11:03,820
other types of mechanisms too?

257
00:11:10,060 --> 00:11:10,580
Cool.

258
00:11:12,460 --> 00:11:15,920
Yeah, I think if that is it, then, oh, do you have a question?

259
00:11:16,340 --> 00:11:16,860
Okay.

260
00:11:22,360 --> 00:11:27,500
How have you seen the kind of the long term maintainability side

261
00:11:27,500 --> 00:11:29,420
of open source projects change?

262
00:11:29,420 --> 00:11:32,980
Especially when it comes down to maybe like resource management

263
00:11:32,980 --> 00:11:37,740
and, you know, bigger, for example, bigger open source projects

264
00:11:37,740 --> 00:11:40,900
have budgets and they can hire people, right?

265
00:11:40,980 --> 00:11:43,880
Or they're sponsored by enterprises.

266
00:11:44,860 --> 00:11:48,360
Yeah, in this context, how has that changed?

267
00:11:49,960 --> 00:11:54,980
Maybe like two initial responses or initial reactions.

268
00:11:54,980 --> 00:11:57,500
I think one is actually related to a conversation I was having with

269
00:11:57,500 --> 00:12:01,520
Max earlier, which is I suspect that we'll probably just need to

270
00:12:01,520 --> 00:12:04,340
be open minded about what contributing to open source means.

271
00:12:05,440 --> 00:12:08,640
Where I think in a previous era, code was the thing, right?

272
00:12:08,740 --> 00:12:11,800
But I think it's pretty clear that code may not be the most valuable

273
00:12:11,800 --> 00:12:12,660
thing that you can contribute.

274
00:12:14,060 --> 00:12:16,660
Arguably, all along, the most valuable thing you can contribute

275
00:12:16,660 --> 00:12:18,060
probably wasn't code to begin with.

276
00:12:18,140 --> 00:12:21,400
It was just like a very easy thing to like kind of gravitate towards.

277
00:12:21,400 --> 00:12:25,160
But really, it was about like creative ideas, your ability to build

278
00:12:25,160 --> 00:12:28,180
trust within an ecosystem, and your ability to build cohesion with

279
00:12:28,180 --> 00:12:29,820
a group of people across the internet.

280
00:12:30,440 --> 00:12:34,360
So I think the nature of contributing and what we choose to value

281
00:12:34,360 --> 00:12:37,700
probably needs to change as well, where code is no longer the valuable

282
00:12:37,700 --> 00:12:37,960
thing.

283
00:12:38,060 --> 00:12:39,800
That's like almost trivially automatable.

284
00:12:39,800 --> 00:12:41,540
So that's like one reaction.

285
00:12:41,900 --> 00:12:46,680
In terms of like how this stuff gets funded, I don't really know.

286
00:12:46,940 --> 00:12:52,060
But I think we already have a lot of funding sources that

287
00:12:52,060 --> 00:12:57,100
have vested interest in having like solid building blocks that they

288
00:12:57,100 --> 00:12:57,800
don't have to reinvent.

289
00:12:57,800 --> 00:13:01,920
And I think this remains true even with agents, where, yeah, you

290
00:13:01,920 --> 00:13:03,520
could totally rebuild these building blocks.

291
00:13:03,940 --> 00:13:06,900
Or you could just glue an existing building block that works really

292
00:13:06,900 --> 00:13:07,200
well.

293
00:13:07,380 --> 00:13:10,140
And I suspect that we'll continue to have that preference going

294
00:13:10,140 --> 00:13:10,460
forward.

295
00:13:10,460 --> 00:13:14,460
And I think, I don't know how it gets funded, but I imagine it'll

296
00:13:14,460 --> 00:13:19,080
be still involving some of the existing corporate sponsors and some

297
00:13:19,080 --> 00:13:19,900
of the existing mechanisms.

298
00:13:20,700 --> 00:13:23,280
But I think what's interesting is that if you assume that a lot

299
00:13:23,280 --> 00:13:27,700
of people have agents and token budgets, how do they allocate, you

300
00:13:27,700 --> 00:13:31,640
know, their money and or now tokens and intelligence towards building

301
00:13:31,640 --> 00:13:32,760
these shared building blocks as well.

302
00:13:32,760 --> 00:13:35,640
So I think that's probably like the open greenfield thing where

303
00:13:35,640 --> 00:13:38,000
that might change the way that we think about how it's quote unquote

304
00:13:38,000 --> 00:13:38,960
funded and sustained.

305
00:13:39,620 --> 00:13:42,300
And I mean, I think part of this talk series is figuring that out.

306
00:13:42,440 --> 00:13:45,940
I don't know the details, but I suspect something new will emerge

307
00:13:45,940 --> 00:13:46,380
there as well.

308
00:13:50,200 --> 00:13:50,600
Cool.

309
00:13:51,280 --> 00:13:52,620
That's it for me.

310
00:13:52,680 --> 00:13:54,520
And I'll pass it off to Max.

311
00:13:55,100 --> 00:13:55,160
Yeah.

312
00:13:56,260 --> 00:13:56,820
Thank you.

313
00:13:57,860 --> 00:13:58,660
Thank you.

314
00:13:58,700 --> 00:13:59,000
you.

315
00:13:59,460 --> 00:13:59,580
Thank you.

316
00:13:59,980 --> 00:14:00,160
Thank

317
00:14:00,400 --> 00:14:00,480
you.